Github user srowen commented on the issue:
https://github.com/apache/spark/pull/14333
Oh, it is indeed building up a lineage. I think it's easier to leave this
broadcast as-is then unless we know that destroying them is essential for
reclaiming driver resources.
Here's another issue though: we now have a lineage where all of the RDDs
are persisted (the cost RDDs), but I think only the last one is unpersisted.
Ideally each would be persisted, materialized, and then unpersist the previous
one. We had this problem and solution pattern in Word2Vec. Does that make sense?
It might be worth tackling here, because, if that's implemented, then I
think the broadcast can safely be disposed at each iteration too, rather than
only at the end.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
